Sampler Scheduler Settings (JPS)

ComfyUI's sampler dropdown has grown to well over thirty options and the scheduler list isn't much shorter, and once you've got more than one KSampler in a graph - a base pass and a hires-fix pass, say, or a couple of comparison branches - re-picking the same sampler/scheduler pair on each one is exactly the kind of busywork that drifts out of sync the first time you forget to update all of them. This node exists so you set that pair once and pipe it everywhere.

It's two dropdowns and two matching outputs. sampler_name covers the full ComfyUI sampler list - euler, the dpmpp family, uni_pc, res_multistep, and everything else currently registered, 34 options deep. scheduler covers the noise-schedule side: normal, karras, exponential, sgm_uniform, simple, ddim_uniform, beta, linear_quadratic, kl_optimal. Pick one of each, and the node outputs them as sampler_name and scheduler - the exact same values, just now available as wires instead of widgets, so you can feed one source into every KSampler that needs to match.

The choice that actually matters here isn't the node, it's what you pick. This is a genuinely easy place to carry stale advice forward. DPM++ 2M Karras was the safe universal default for years, but that's specifically an SD 1.5/SDXL-era recommendation - it belongs to DDPM-style models (SDXL and its finetunes: Illustrious, NoobAI, Pony) where a reshaped noise curve corrects a curved denoising trajectory. Flow-matching architectures (Flux, Z-Image, and their relatives) train on a near-straight trajectory instead, and community sweeps have found Karras and exponential schedulers genuinely fail on those models rather than merely underperform - not a taste difference, a broken combination. If you're on SDXL, DPM++ 2M Karras (or Euler a for anime finetunes) is still the right call; if you're on a flow-matching model, reach for Euler-family samplers on beta or simple instead, and don't carry the Karras habit across architectures out of muscle memory.

Where it fits. Wire sampler_name and scheduler into every KSampler (or KSampler Advanced) that should share the same settings. The README calls this out directly: it "can also be used as pipe," meaning its real value shows up once you've got more than one sampling node in the graph and want a single source of truth for both.

Installing it. ComfyUI Manager, search "JPS Custom Nodes for ComfyUI," or manually:

cd ComfyUI/custom_nodes
git clone https://github.com/JPS-GER/ComfyUI_JPS-Nodes.git

Restart and it's under JPS Nodes/Settings. Delete any older copy of the pack first if reinstalling. No models, no extra dependencies - it's just exposing ComfyUI's own sampler/scheduler lists as node outputs.

Troubleshooting. If a sampler or scheduler you expect isn't in the dropdown, that's almost always a version mismatch - this node reflects whatever your ComfyUI install currently has registered, so a sampler added in a recent ComfyUI update won't show up here until you've updated ComfyUI itself, and a custom sampler pack (like RES4LYF) that adds its own solvers won't appear in this node's list unless it hooks into the same registration point. This node only exposes the stock list. If your workflow depends on a third-party sampler, you'll need that pack's own node instead of trying to select it here.